Training Factory open-sources QLoRA fine-tuning on 4-bit CQ-quantized models
textclf · reddit · 2026-10-21
A Reddit user released Training Factory, an open-source platform for fine-tuning 4-bit quantized open models via QLoRA to cut GPU costs. It uses a custom calibration-free quantization scheme called CQ (Coded Quantization), claimed to match leading quant methods as a base.
The post walks through a full five-step workflow: install from GitHub, download a CQ 4-bit base (e.g. Qwen3.8-27B-CQ-4bit), train a LoRA adapter targeting qproj/vproj, serve base + adapter with vLLM (--quantization cq --enable-lora), and query the fine-tuned adapter via the chat completions API. Available bases include Qwen3.8-27B, Llama-3.1-8B-Instruct, Llama-3.3-70B-Instruct and Qwen3-Coder-Next.
More from coding & agent
- OpenAI: Two Internal Coding Agents Weakened Their Own Misalignment Monitoring — rohanpaul_ai · 2026-10-21
- Tokyo DevDay: strict enterprise plugin policies push devs toward private plugins — sierracatalina · 2026-10-21
- New agent stack: Grok Bot orchestrates Gemini, ChatGPT and Claude via CLIs — _AustinCalvert_ · 2026-10-21
- Dev builds attest: position-aware prompt-injection scanner for MCP tool descriptions — mrpkk · 2026-10-21
- S-Agent 2.0 turns coding agents into spatial reasoners, hitting 66.5 on MMSI-Bench — liuziwei7 · 2026-10-21
- LLM trading agent skips 380 of 392 calls — scoring refusals shows skipped events moved 2x random — Tasty-Mammoth5321 · 2026-10-21